Generalized labeled multi-Bernoulli space-object tracking with joint prediction and update

Abstract

Space-object tracking systems require robust and accurate methods of multi-target state estimation and prediction. This paper presents the application of labeled multi-Bernoulli filters for space-object tracking, and leverages a joint prediction and update with Gibbs sampling to improve computational efficiency. Based on the use of labeled random finite sets, the d-Generalized Labeled Multi-Bernoulli Filter provides a closed-form solution to the Bayes recursion for a multi-target filter. A similar filter, the Labeled Multi-Bernoulli Filter, is a principled approximation to reduce computational complexity. Upon combining these filters with astrodynamics-based models for orbit state probability density function prediction and initial orbit determination, a 100-object simulation is used to demonstrate the ability of these tools to track space objects in near-geosynchronous orbit. Both filters converge on solutions with sub-500 meter accuracy and demonstrate similar performance as a function of detection probability, clutter, and the birth model employed. A robust comparison of the two filters requires further Monte Carlo-based tests to quantify variance in the solutions due to random inputs.